Convolutional neural networks (CNNs) have been widely used in different areas. The success of CNNs comes with a huge amount of parameters and computations, and nowaday CNNs still keep moving toward larger structures. ...
详细信息
ISBN:
(纸本)9781450394451
Convolutional neural networks (CNNs) have been widely used in different areas. The success of CNNs comes with a huge amount of parameters and computations, and nowaday CNNs still keep moving toward larger structures. Although larger structures often bring about better inference accuracy, the increasing size also slows the inference speed down. Recently, various parameter sparsity methods have been proposed to accelerate CNNs by reducing the number of parameters and computations. Existing sparsity methods could be classified into two categories: unstructured and structured. Unstructured sparsity methods easily cause irregularity and thus have a suboptimal speedup. On the other hand, the structured sparsity methods could keep regularity by pruning the parameters following a certain pattern but result in low sparsity. In this paper, we propose a software/hardware co-design approach to bring local irregular sparsity into CNNs. Benefiting from the local irregularity, we design a row-wise computing engine, RConv Engine, to achieve workload balance and remarkable speedup. The experimental results show that our software/hardware co-design method can achieve a 10.9x speedup than the state-of-the-art methods with a negligible accuracy loss.
Driven by the potential applications of ionic liquid(IL)flow for charging graphene-based surfaces in many emerging technologies,recent research efforts have focused on understanding ion dynamics and structuring at IL-...
详细信息
Driven by the potential applications of ionic liquid(IL)flow for charging graphene-based surfaces in many emerging technologies,recent research efforts have focused on understanding ion dynamics and structuring at IL-graphene ***,graphene colloid probe(GrP)atomic force microscopy(AFM)was used to probe the dynamics and ion structuring of 1-butyl-3-methylimidazolium tetrafluoroborate at graphene surfaces under various bias *** particular,the AFM-measured nanofriction provides a good measure of the dynamic properties of the ILs at graphene *** with the IL at the unbiased graphene surface(0 V),the charged graphene surfaces with either negative(-1,-2 V)or positive(+1,+2 V)voltages favor a reduction in the friction coefficient by the IL.A higher magnitude of the bias voltage applied on the graphene surface with either sign(-2 or+2 V)results in a smaller friction coefficient than that at-1 and+1 *** combination with the AFM-probed contact stiffness,adhesion forces,and ion structuring force curves with an ion orientational distribution according to molecular dynamics(MD)simulations,we discovered that the unbiased graphene surface(0 V)possesses randomly structured IL ions and that the graphene colloid probe is more likely to become stuck,resulting in more energy dissipation to contribute to a larger friction *** of the graphene surface under either negative or positive voltages resulted in uniformly arranged ions,which produced a more ordered ion structure and,thus,a smoother sliding plane to reduce the friction *** impedance spectroscopy(EIS)for the IL with graphene as an electrode demonstrated a greater ionic conductivity in the IL paired with the biased graphene than in the unbiased one,implying faster ion movement at the charged graphene,which is beneficial for reducing the friction coefficient.
An important goal in synthetic biology is in the design and implementation of control strategies embedded in biochemical reaction systems to enable control of the process at the molecular level. The problem of set-poi...
详细信息
With the continuous development of deep learning, deep neural networks are gradually applied to traffic classification problems. However, the large network structure and parameter number of deep neural networks hinder...
详细信息
ISBN:
(纸本)9798400700132
With the continuous development of deep learning, deep neural networks are gradually applied to traffic classification problems. However, the large network structure and parameter number of deep neural networks hinder the application on edge computing devices. Reducing network scale helps relieve computational pressure, this paper proposes a lightweight traffic classification model to provide reliable accuracy and reduce the consumption of computing resources. In this work, we design an F1DCN network, which takes full advantage of the convolution layer parameters and the convolution kernel field of view. The lightweight approach effectively improves the classification performance and saves massive parameters. The model pruning method is applied to find the optimal structure of the network. Experiments on two public datasets show that the proposed model reduce more than 80 % parameters and 45 % FLOPS compared with traditional traffic classification methods, and maintaining more than 95 % classification accuracy.
In order to achieve better effect on blocking virus propagation, a neighbor immunization strategy was proposed and verified in this paper. On email-EU real networks and on scale-free model network, the efficiencies of...
详细信息
SM4-GCM is an encryption algorithm with authentication function. The algorithm achieves the purpose of data security and information integrity. The SM4-GCM algorithm, implemented using traditional software methods, ha...
详细信息
The spatial diversity and multiplexing advantages of massive multi-input-multi-output (mMIMO) can significantly improve the capacity of massive non-orthogonal multiple access (NOMA) in machine type communications. How...
详细信息
Concurrent B+trees have been widely used in many systems. With the scale of data requests increasing exponentially, the systems are facing tremendous performance pressure. GPU has shown its potential to accelerate con...
详细信息
Point-based methods have made significant progress, but improving their scalability in large-scale 3D scenes is still a challenging problem. In this paper, we delve into the point-based method and develop a simpler, f...
As the ultimate goal of steganalysis, secret message extraction is a bottleneck and difficulty that has long plagued the development of steganalysis technology. Existing pioneering work on secret message extraction fo...
详细信息
暂无评论